UGC Activewear Video Generator

Generate TikTok / Instagram Reels style activewear UGC product videos from a single outfit image.

Requires: imastudio-cli npm package (ima command) and IMA_API_KEY. Get your API key at: https://imastudio.com

Workflow

When the user provides an outfit image (local file or URL), execute these steps in order:

Step 1 — Analyze the Image

Use your vision capability to extract from the image:

• Model: gender, body type, vibe, hairstyle
• Outfit: type (set/top/bottom), silhouette, fabric texture, color palette
• Details: logo placement, accessories, fit and proportions
• Environment: studio / outdoor / gym / street

Step 2 — Lock Consistency Rules

Every generated video MUST maintain across all shots:

• Same face, body type, hairstyle, makeup
• Same outfit with exact color, silhouette, fit, logo placement
• Same accessories and styling energy

Tell the model explicitly in every prompt: "same model, same outfit, same styling throughout."

Step 3 — Generate Two Video Prompts

Build two 15-second video prompts:

A) Talking Head (influencer speaks to camera):

• 0–3s: Hook shot — direct-to-camera line, outfit visible
• 3–7s: Detail close-ups — fabric, waistband, body-line flattery
• 7–11s: Movement — walking, turning, lifestyle motion
• 11–15s: Hero shot — full-body, closing line

B) Voiceover (aesthetic b-roll + narration):

• 0–3s: Outfit entrance — hero reveal shot
• 3–7s: Detail shots — texture, silhouette, comfort cues
• 7–11s: Lifestyle motion — natural movement, posing
• 11–15s: Full-body hero + CTA

Visual rules for both: handheld but polished, punch-in zooms, natural daylight, clean transitions, strong silhouette emphasis, premium social-native look.

Step 4 — Upload Image (if local file)

ima upload <outfit-image> --json

Use the returned url as input for video generation.

Step 5 — Generate Videos

For each prompt (talking head + voiceover), run:

ima create-task \
  --task-type image_to_video \
  --model wan2.6-i2v \
  --param prompt="<assembled prompt>" \
  --param input_images="<image_url>" \
  --param duration=10 \
  --param aspect_ratio=9:16 \
  --wait --json

Model selection:

Use 9:16 aspect ratio (vertical/portrait) for TikTok and Reels.

Step 6 — Generate TTS Narration (Voiceover only)

For the voiceover video, generate a spoken script:

ima create-task \
  --task-type text_to_speech \
  --model seed-tts-1.1 \
  --param prompt="<voiceover script>" \
  --wait --json

Script tone: confident, aspirational, social-native. Not salesy — like a friend recommending a find.

Step 7 — Deliver Results

Send each video to the user with:

• Video via media URL (inline playback)
• Which style it is (talking head / voiceover)
• Model used and generation time
• The prompt and script used (so they can iterate)

Prompt Assembly Guide

When building the video generation prompt, include ALL of these elements:

1. Scene setup: "A [gender] fitness influencer in [location], [lighting]"
2. Outfit description: exact details from Step 1 analysis
3. Action sequence: what happens in each time segment
4. Camera work: "handheld, slight movement, punch-in zoom on [detail]"
5. Mood/energy: "confident, premium athleisure, social-media-native"
6. Consistency anchor: "same model, same outfit, same styling throughout the video"

Example prompt:

A confident young woman in a modern minimalist apartment, natural daylight. She wears a matching sage-green ribbed sports bra and high-waisted leggings set with subtle logo on waistband. She looks at camera with a warm smile, then the camera punches in on the fabric texture and waistband detail. She turns showing the silhouette from the side, walks toward the window. Final full-body hero shot, hands on hips. Handheld camera, premium social-media look. Same model, same outfit, same styling throughout.

Script Templates

Talking head hook lines:

• "Okay this set is actually insane"
• "POV: you found the perfect gym-to-brunch set"
• "I need everyone to see this fabric up close"
• "This might be my new favorite workout set"

Voiceover narration example:

"When I say this set hits different — I mean it. The ribbing, the compression, the way it moves with you. From the gym to coffee runs, this is the one."

Input Parameters

Notes

• Always use image_to_video (not text_to_video) to maintain outfit consistency from the source image
• Vertical 9:16 is default — only use 16:9 if user explicitly asks for landscape
• If the first result has consistency issues, retry with kling-video-o1 which has stronger reference adherence
• For batch production (multiple outfits), process one at a time and deliver incrementally